Support internal function to_binary() for new charset

[tangenta]

The code changes about built-in functions in TiKV can be minimized.

After the investigation about several built-in functions, I found some common patterns:

1. Create a new encoding.
2. Convert the arguments into new charset from utf8.
3. Apply these changes to the vectorized-version evaluation.

We can extract these patterns into another Sig named builtinToBinarySig(to_binary()) and wrap the cast function as follows:

The rewrite process before this proposal:

select some_builtin(arg1, arg2...)

is rewritten to

select some_builtin(cast_as_string(arg1), cast_as_string(arg2)...)

The rewrite process after this proposal:

select some_builtin(arg1, arg2...)

is rewritten to

select some_builtin(to_binary(cast_as_string(arg1)), to_binary(cast_as_string(arg2))...)

Then we can maintain a built-in function list to determine which function needs to be wrapped, instead of putting the boilerplates everywhere.

var toBinaryMap = map[string]struct{}{
	ast.Hex: {}, ast.Length: {}, ast.OctetLength: {}, ast.ASCII: {},
	ast.ToBase64: {},
}

[tangenta]

@zimulala @xiongjiwei @ou-bing @Defined2014 @unconsolable @jackwener @chacha923 @hawkingrei @jayl-zxl FYI, please feel free to leave your suggestions here, if any.

[unconsolable]

Looks good since it can reduce code repetition, but I have a small question.
This code snippet comes from (*builtinInternalToBinarySig).vecEvalString

	for i := 0; i < n; i++ {
		var str string
		if buf.IsNull(i) {
			result.AppendNull()
			continue
		}
		str = buf.GetString(i)
		str, err = enc.EncodeString(str)

I wonder whether an encode buffer should be introduced here to reduce memory allocation.

[tangenta]

@unconsolable Good catch! If you are interested in it you can file a PR to fix it.

By the way, we are trying to collect all the functions that can be wrapped by to_binary() and add them to the list, you can also include some of them in your PR.